Outline. Vorlesung Telematik. Higher Layer Protocols: TCP/IP and ATM. Packet Switching. Higher Layers. Network Layer and Internetworking

Transcription

1 Outline Network Layer and Internetworking Vorlesung Telematik Higher Layer Protocols: TCP/IP and ATM The TCP/IP protocol suit ATM MPLS Xiaoming Fu Telematics Group University of Goettingen Most slides were adopted from Eytan Modiano, MIT; some are adapted from Jim Kurose & Keith Ross "Computer Networking" (2nd ed) Slide 2 Higher Layers Packet Switching Application Application Virtual network service Presentation Presentation Virtual session Session Session Virtual link for end to end messages Transport Transport Virtual link for end to end packets Network Network Network Network Virtual link for reliable packets Data link Data link Control DLC DLC DLC DLC Control Virtual bit pipe physical physical interface phys. int. phys. int. phys. int. phys. int. interface Physical link External subnet subnet External Site node node site TCP, UDP IP, ATM Datagram packet switching Route chosen on packet-by-packet basis Different packets may follow different routes Packets may arrive out of order at the destination E.g., IP (The Internet Protocol) Virtual Circuit packet switching All packets associated with a session follow the same path Route is chosen at start of session Packets are labeled with a VC# designating the route The VC number must be unique on a given link but can change from link to link Imagine having to set up connections between nodes in a mesh Unique VC numbers imply Million VC numbers that must be represented and stored at each node E.g., ATM (Asynchronous transfer mode) Slide 3 Slide 4

2 Virtual Circuits Packet Switching The TCP/IP Protocol Suite For datagrams, addressing information must uniquely distinguish each network node and session Need unique source and destination addresses Transmission Control Protocol / Internet Protocol For virtual circuits, only the virtual circuits on a link need be distinguished by addressing Global address needed to set-up virtual circuit Once established, local virtual circuit numbers can then be used to represent the virtual circuits on a given link: VC number changes from link to link Merits of virtual circuits Save on route computation Need only be done once at start of session Save on header size More complex Less flexible 3 6 VC3 VC7 VC3 VC4 5 8 VC3 VC7 Node 5 table (3,5) VC3 -> (5,8) VC3 (3,5) VC7 -> (5,8) VC4 (6,5) VC3 -> (5,8) VC7 2 9 Developed by DARPA to connect Universities and Research Labs Four Layer model Applications Transport Network Link TCP - Transmission Control Protocol UDP - User Datagram Protocol IP - Internet Protocol Telnet, FTP, , etc. TCP, UDP IP, ICMP, IGMP Device drivers, interface cards Slide 5 Slide 6 Internetworking with TCP/IP Encapsulation FTP FTP Protocol FTP client server TCP TCP Protocol ROUTER IP IP Protocol IP Protocol IP Ethernet driver Ethernet Protocol Ethernet Ethernet driver token ring driver token ring Protocol token ring TCP IP token ring driver IP header TCP header TCP header IP datagram Appl header user data user data application data TCP segment application data Ethernet IP TCP Ethernet header header header application data trailer Ethernet frame 46 to 5 bytes Application TCP IP Ethernet driver Ethernet Slide 7 Slide 8

3 Bridges, Routers and Gateways Bridges, routers and gateways A Bridge is used to connect multiple LAN segments Layer 2 routing (Ethernet) Does not know IP address Varying levels of sophistication Simple bridges just forward packets smart bridges start looking like routers A Router is used to route connect between different networks using network layer address Within or between Autonomous Systems Using same protocol (e.g., IP, ATM) A Gateway connects between networks using different protocols Protocol conversion Address resolution Small company Ethernet A Bridge Ethernet B IP Router Gateway Service provider s ATM backbone Another provider s Frame Relay Backbone Gateway These definitions are often mixed and seem to evolve! ATM switches (routers) Slide 9 Slide IP addresses Host Names 32 bit address written as four decimal numbers One per byte of address (e.g., ) Hierarchical address structure Network ID/ Host ID/ Port ID Complete address called a socket Network and host ID carried in IP Header Port ID (sending process) carried in TCP header Each machine also has a unique name Domain name System: A distributed database that provides a mapping between IP addresses and Host names E.g., => IP Address classes: 8 32 Net ID Host ID Class A Nets 6 32 Net ID Host ID Class B Nets Net ID Host ID Class C Nets Class D is for multicast traffic Slide Slide 2

4 Internet Standards The Internet Protocol (IP) Internet Engineering Task Force (IETF) Development on near term internet standards Open body Meets 3 times a year Request for Comments (RFCs) Official internet standards Available from IETF web page: Routing of packet across the network Unreliable service Best effort delivery Recovery from lost packets must be done at higher layers Connectionless Packets are delivered (routed) independently Can be delivered out of order Re-sequencing must be done at higher layers Current version: V4 Next generation: V6 Add more addresses (4 byte header!) Ability to provide QoS and better security Slide 3 Slide 4 Header Fields in IP IP HEADER FIELDS Header Ver length type of service Total length (bytes) 6 - bit identification Flags 3 - bit fragment offset TTL Protocol Source IP Address Destination IP Address Options (if any) Data Header Checksum Note that the minimum size header is 2 bytes; TCP also has 2 byte header Vers: Version # of IP (current version is 4) HL: Header Length in 32-bit words Service: Mostly Ignored Total length Length of IP datagram ID Unique datagram ID Flags: NoFrag, More FragOffset: Fragment offset in units of 8 Octets TTL: Time to Live in "seconds or Hops Protocol: Higher Layer Protocol ID # HDR Cksum: 6 bit 's complement checksum (on header only!) SA & DA: Network Addresses Options: Record Route,Source Route,TimeStamp Slide 5 Slide 6

5 FRAGMENTATION POSITION OF FRAGMENT ethernet mtu=5 G X.25 MTU = 52 G ethernet mtu=5 Fragment offset field gives starting position of fragment within datagram in 8 byte increments (3 bit field) Length field in header gives the total length in bytes (6 bit field) Maximum size of IP packet 64K bytes A gateway fragments a datagram if length is too great for next network (fragmentation required because of unknown paths). Each fragment needs a unique identifier for datagram plus identifier for position within datagram A flag bit denotes last fragment in datagram IP reassembles fragments at destination and throws them away if one or more is too late in arriving In IP, the datagram ID is a 6 bit field counting datagram from given host Slide 7 Slide 8 IP Routing Subnet addressing Routing table at each node contains for each destination the next hop router to which the packet should be sent Not all destination addresses are in the routing table Look for net ID of the destination Prefix match Use default router Routers do not compute the complete route to the destination but only the next hop router IP uses distributed routing algorithms: RIP, OSPF In a LAN, the host computer sends the packet to the default router which provides a gateway to the outside world Class A and B addresses allocate too many hosts to a given net Subnet addressing allows us to divide the host ID space into smaller sub networks Simplify routing within an organization Smaller routing tables Potentially allows the allocation of the same class B address to more than one organization 32 bit Subnet Mask is used to divide the host ID field into subnets denotes a network address field denotes a host ID field 6 bit net ID 6 bit host ID Class B Address Subnet ID Host ID Mask Slide 9 Slide 2

6 Classless inter-domain routing (CIDR) Dynamic Host Configuration (DHCP) Class A and B addresses allocate too many hosts to an organization while class C addresses don t allocate enough This leads to inefficient assignment of address space Classless routing allows the allocation of addresses outside of class boundaries (within the class C pool of addresses) Allocate a block of contiguous addresses E.g., Bundles 6 class C addresses The first 2 bits of the address field are the same and are essentially the network ID Network numbers must now be described using their length and value (I.e., length of network prefix) Routing table lookup using longest prefix match Automated method for assigning network numbers IP addresses, default routers Computers contact DHCP server at Boot-up time Server assigns IP address Allows sharing of address space More efficient use of address space Adds scalability Addresses are least for some time Not permanently assigned Notice similarity to subnetting - supernetting Slide 2 Slide 22 Address Resolution Protocol IP addresses only make sense within IP suite Local area networks, such as Ethernet, have their own addressing scheme To talk to a node on LAN one must have its physical address (physical interface cards don t recognize their IP addresses) ARP provides a mapping between IP addresses and LAN addresses RARP provides mapping from LAN addresses to IP addresses This is accomplished by sending a broadcast packet requesting the owner of the IP address to respond with their physical address All nodes on the LAN recognize the broadcast message The owner of the IP address responds with its physical address An ARP cache is maintained at each node with recent mappings ARP IP Ethernet RARP Routing in the Internet The internet is divided into sub-networks, each under the control of a single authority known as an Autonomous System (AS) Routing algorithms are divided into two categories: Interior protocols (within an AS) Exterior protocols (between AS s) Interior Protocols use shortest path algorithms (more later) Distance vector protocols based on Bellman-ford algorithm Nodes exchange routing tables with each other E.g., Routing Information Protocol (RIP) Link state protocols based on Dijkstra s algorithm Nodes monitor the state of their links (e.g., delay) Nodes broadcast this information to all of the network E.g., Open Shortest Path First (OSPF) Exterior protocols route packets across AS s Issues: no single cost metric, policy routing, etc.. Routes often are pre-computed Example protocols: Exterior Gateway protocol (EGP) and Border Gateway protocol (BGP) Slide 23 Slide 24

7 IPv6 Effort started in 99 as IPng Motivation Need to increase IP address space Support for real time application - QoS Security, Mobility, Auto-configuration Major changes Increased address space (6 bytes) 5 IP addresses per sq. ft. of earth! Address partition similar to CIDR Support for QoS via Flow Label field Simplified header Most of the reasons for IPv6 have been taken care of in IPv4 Is IPv6 really needed? Complex transition from V4 to V6 3 ver class Flow label length nexthd Hop limit Source address Destination address Network Address Translation (NAT) Insufficient IP addresses for many organizations IPv6: but most applications run over IPv4 "Transient" solution: NAT Only one address can allow for a large amount of hosts Need to maintain a mapping <port #, private IP address> at NAT gateways NAT pros & cons advantages: no need to be allocated range of addresses from ISP; can change a device's address without notifying outside world; can change ISP without changing local devices' addresses; local devices are not addressable from outside Controversal: Routers should only process up to layer 3 (violates the "end2end" argument) Address shortage should be instead resolved by IPv6 Slide 25 Slide 26 User Datagram Protocol (UDP) NAT: Network Address Translation 2: NAT router changes datagram source addr from..., 3345 to , 5, updates table S: , 8 D: , 5 3 3: Reply arrives dest. address: , 5 Slide 27 NAT translation table WAN side addr LAN side addr , 5..., 3345 S: , 5 D: , S:..., 3345 D: , 8 S: , 8 D:..., : host... sends datagram to , : NAT router changes datagram dest addr from , 5 to..., 3345 Transport layer protocol Delivery of messages across network Datagram oriented Unreliable No error control mechanism Connectionless Not a stream protocol Max packet length 65K bytes UDP checksum Covers header and data Optional Can be used by applications UDP allows applications to interface directly to IP with minimal additional processing or protocol overhead Slide 28

8 UDP header format Transmission Control Protocol (TCP) IP Datagram IP header UDP header data 6 bit source port number 6 bit destination port number 6 bit UDP length 6 bit checksum Data The port numbers identifie the sending and receiving processes I.e., FTP, , etc.. Allow UDP to multiplex the data onto a single stream UDP length = length of packet in bytes Minimum of 8 and maximum of 2^6 - = 65,535 bytes Checksum covers header and data Optional, UDP does not do anything with the checksum Transport layer protocol Reliable transmission of messages Connection oriented Stream traffic Must re-sequence out of order IP packets Reliable ARQ mechanism Notice that packets have a sequence number and an ack number Notice that packet header has a window size (for Go Back N) Flow control mechanism Slow start Limits the size of the window in response to congestion Slide 29 Slide 3 Basic TCP operation TCP header fields At sender Application data is broken into TCP segments TCP uses a timer while waiting for an ACK of every packet Un-ACK d packets are retransmitted Source port Sequence number 6 32 Destination port At receiver Errors are detected using a checksum Correctly received data is acknowledged Segments are reassembled into their proper order Duplicate segments are discarded Window based retransmission and flow control Request number Data Offset Reserved Control Window Check sum Urgent pointer Options (if any) Data Slide 3 Slide 32

9 TCP header fields TCP error recovery Ports number are the same as for UDP 32 bit SN uniquely identify the application data contained in the TCP segment SN is in bytes! It identify the first byte of data 32 bit RN is used for piggybacking ACK s RN indicates the next byte that the received is expecting Implicit ACK for all of the bytes up to that point Data offset is a header length in 32 bit words (minimum 2 bytes) Window size Used for error recovery (ARQ) and as a flow control mechanism Sender cannot have more than a window of packets in the network simultaneously Specified in bytes Window scaling used to increase the window size in high speed networks Checksum covers the header and data Error recovery is done at multiple layers Link, transport, application Transport layer error recovery is needed because Packet losses can occur at network layer E.g., buffer overflow Some link layers may not be reliable SN and RN are used for error recovery in a similar way to Go Back N at the link layer Large SN needed for re-sequencing out of order packets TCP uses a timeout mechanism for packet retransmission Timeout calculation Fast retransmission Slide 33 Slide 34 TCP timeout calculation Fast Retransmit Based on round trip time measurement (RTT) Weighted average RTT_AVE = a*(rtt_measured) + (-a)*rtt_ave Timeout is a multiple of RTT_AVE (usually two) Short Timeout would lead to too many retransmissions Long Timeout would lead to large delays and inefficiency In order to make Timeout be more tolerant of delay variations it has been proposed (Jacobson) to set the timeout value based on the standard deviation of RTT Timeout = RTT_AVE + 4*RTT_SD In many TCP implementations the minimum value of Timeout is 5 ms due to the clock granularity When TCP receives a packet with a SN that is greater than the expected SN, it sends an ACK packet with a request number of the expected packet SN This could be due to out-of-order delivery or packet loss If a packet is lost then duplicate RNs will be sent by TCP until the packet it correctly received But the packet will not be retransmitted until a Timeout occurs This leads to added delay and inefficiency Fast retransmit assumes that if 3 duplicate RNs are received by the sending module that the packet was lost After 3 duplicate RNs are received the packet is retransmitted After retransmission, continue to send new data Fast retransmit allows TCP retransmission to behave more like Selective repeat ARQ Future option for selective ACKs (SACK) Slide 35 Slide 36

10 TCP congestion control Effect Of Window Size TCP uses its window size to perform end-to-end congestion control More on window flow control later Basic idea With window based ARQ the number of packets in the network cannot exceed the window size (CW) The window size is the number of bytes that are allowed to be in transport simultaneously WINDOW WASTED BW WINDOW Last_byte_sent (SN) - last_byte_ack d (RN) <= CW Transmission rate when using window flow control is equal to one window of packets every round trip time R = CW/RTT By controlling the window size TCP effectively controls the rate Too small a window prevents continuous transmission To allow continuous transmission window size must exceed round-trip delay time Slide 37 Slide 38 Length of a bit (traveling at 2/3C) Dynamic adjustment of window size At 3 bps bit = 45 miles 3 miles = 7 bits At 3.3 kbps bit = 38 miles 3 miles = 79 bits At 56 kbps bit = 2 miles 3 miles =.5 kbits At.5 Mbps bit = 438 ft. 3 miles = 36 kbits At 5 Mbps bit = 4.4 ft. 3 miles = 3.6 Mbits At Gbps bit = 8 inches 3 miles = 24 Mbits TCP starts with CW = packet and increases the window size slowly as ACK s are received Slow start phase Congestion avoidance phase Slow start phase During slow start TCP increases the window by one packet for every ACK that is received When CW = Threshold TCP goes to Congestion avoidance phase Notice: during slow start CW doubles every round trip time Exponential increase! Congestion avoidance phase During congestion avoidance TCP increases the window by one packet for every window of ACKs that it receives Notice that during congestion avoidance CW increases by every round trip time - Linear increase! TCP continues to increase CW until congestion occurs Slide 39 Slide 4

11 Reaction to congestion Understanding TCP dynamics Many variations: Tahoe, Reno, Vegas Basic idea: when congestion occurs decrease the window size There are two congestion indication mechanisms Duplicate ACKs - could be due to temporary congestion Timeout - more likely due to significant congstion TCP Reno - most common implementation Slow start phase is actually fast TCP spends most of its time in Congestion avoidance phase While in Congestion avoidance CW increases by every RTT CW decreases by a factor of two with every loss Additive Increase / Multiplicative decrease If Timeout occurs, CW = and go back to slow start phase If duplicate ACKs occur CW = CW/2 stay in congestion avoidance phase CW Saw-tooth Behavior Time Slide 4 Slide 42 Random Early Detection (RED) TCP Error Control Instead of dropping packet on queue overflow, drop them probabilistically earlier Motivation Dropped packets are used as a mechanism to force the source to slow down If we wait for buffer overflow it is in fact too late and we may have to drop many packets Leads to TCP synchronization problem where all sources slow down simultaneously RED provides an early indication of congestion Randomization reduces the TCP synchronization problem Mechanism Use weighted average queue size If AVE_Q > T min drop with prob. P If AVE_Q > T max drop with prob. RED can be used with explicit congestion notification rather than packet dropping RED has a fairness property Large flows more likely to be dropped Threshold and drop probability values are an area of active research P max T min Ave queue length T max SRP GO BACK N EFFICIENCY VS. BER WITH TCP WINDOW CONSTRAINT E-7 E-6 E-5 E-4 E-3 E-2 CHANNEL BER SEC R/T DELAY T- RATE BIT PACKETS Original TCP designed for low BER, low delay links Future versions (RFC 323) will allow for larger windows and selective retransmissions Slide 43 Slide 44

12 Impact of transmission errors on TCP congestion control TCP releases EFFICIENCY VS BER FOR TCP'S CONGESTION CONTROL KBPS KBPS,544 KBPS 64 KBPS..E-7.E-6.E-5.E-4.E-3 BER TCP assumes dropped packets are due to congestion and responds by reducing the transmission rate Over a high BER link dropped packets are more likely to be due to errors than to congestion TCP extensions (RFC 323) Fast retransmit mechanism, fast recovery, window scaling TCP standards are published as RFC s TCP implementations sometimes differ from one another May not implement the latest extensions, bugs, etc. The de facto standard implementation is BSD Computer system Research group at UC-Berkeley Most implementations of TCP are based on the BSD implementations SUN, MS, etc. BSD releases 4.2BSD First widely available release 4.3BSD Tahoe Slow start and congestion avoidance 4.3BSD Reno - 99 Header compression 4.4BSD Multicast support, RFC 323 for high performance Slide 45 Slide 46 The TCP/IP Suite Asynchronous Transfer Mode (ATM) ping Telnet& FTP Trace SMTP X DNS TFTP BOOTP SNMP Rlogin route TCP ICMP IP UDP IGMP RPC NFS 98 s effort by the phone companies to develop an integrated network standard (BISDN) that can support voice, data, video, etc. ATM uses small (53 Bytes) fixed size packets called cells Why cells? Cell switching has properties of both packet and circuit switching Easier to implement high speed switches Why 53 bytes? Small cells are good for voice traffic (limit sampling delays) For 64Kbps voice it takes 6 ms to fill a cell with data ATM networks are connection oriented Virtual circuits ARP Data Link RARP media Slide 47 Slide 48

13 ATM Reference Architecture ATM Cell format 5 Bytes 48 Bytes Upper layers Applications TCP/IP ATM adaptation layer Similar to transport layer Provides interface between upper layers and ATM Break messages into cells and reassemble ATM layer Cell switching Congestion control Physical layer ATM designed for SONET Synchronous optical network TDMA transmission scheme with 25 s frames Upper Layers ATM Adaptation Layer (AAL) ATM Physical ATM Cell Header Data Virtual circuit numbers (notice relatively small address space!) Virtual channel ID Virtual path ID PTI - payload type CLP - cell loss priority ( bit!) Mark cells that can be dropped HEC - CRC on header ATM Header (NNI) VPI 2 VPI VCI 3 VCI 4 5 VCI PTI HEC C L P Slide 49 Slide 5 VPI/VCI ATM HEADER CRC VPI identifies a physical path between the source and destination VCI identifies a logical connection (session) within that path Approach allows for smaller routing tables and simplifies route computation Private network ATM Backbone ATM uses an 8 bit CRC that is able to correct error It checks only on the header of the cell, and alternates between two modes In detection mode it does not correct any errors but is able to detect more errors In correction mode it can correct up to one error reliably but is less able to detect errors When the channel is relatively good it makes sense to be in correction mode, however when the channel is bad you want to be in detection mode to maximize the detection capability Private network Use VPI for switching in backbone Private network Use VCI to ID connection Within private network No detected errors Correct errors Detect double error Correct single error No detected errors Detect errors Detected errors Slide 5 Slide 52

14 ATM Service Categories ATM service parameters (examples) Constant Bit Rate (CBR) - e.g. uncompressed voice Circuit emulation Variable Bit Rate (rt-vbr) - e.g. compressed video Real-time and non-real-time Available Bit Rate (ABR) - e.g. LAN interconnect For bursty traffic with limited BW guarantees and congestion control Unspecified Bit Rate (UBR) - e.g. Internet ABR without BW guarantees and congestion control Peak cell rate (PCR) Sustained cell rate (SCR) Maximum Burst Size (MBS) Minimum cell rate (MCR) Cell loss rate (CLR) Cell transmission delay (CTD) Cell delay variation (CDV) Not all parameters apply to all service categories E.g., CBR specifies PCR and CDV VBR specifies MBR and SCR Network guarantees QoS provided that the user conforms to his contract as specified by above parameters When users exceed their rate network can drop those packets Cell rate can be controlled using rate control scheme (leaky bucket) Slide 53 Slide 54 Flow control in ATM networks (ABR) End-to-End RM-Cell Flow ATM uses resource management cells to control rate parameters Forward resource management (FRM) Backward resource management (BRM) RM cells contain Congestion indicator (CI) No increase Indicator (NI) Explicit cell rate (ER) Current cell rate (CCR) Min cell rate (MCR) Source generates RM cells regularly As RM cells pass through the networked they can be marked with CI= to indicate congestion RM cells are returned back to the source where CI = => decrease rate by some fraction CI = => Increase rate by some fraction ER can be used to set explicit rate ABR Source D BRM D D = data cell FRM = forward RM cell BRM = backward RM cell ABR Switch D FRM D ABR Switch At the destination the RM cell is turned around and sent back to the source D BRM FRM ABR Destin ation Slide 55 Slide 56

15 ATM Adaptation Layers Interface between ATM layer and higher layer packets Four adaptation layers that closely correspond to ATM s service classes AAL- to support CBR traffic AAL-2 to support VBR traffic AAL-3/4 to support bursty data traffic AAL-5 to support IP with minimal overhead The functions and format of the adaptation layer depend on the class of service. For example, stream type traffic requires sequence numbers to identify which cells have been dropped. Example: AAL 3/4 ATM CELL PAYLOAD (48 Bytes) ST SEQ MID LEN CRC Byte User Payload ST: Segment Type (st, Middle, Last) SEQ:4-bit sequence number (detect lost cells) MID: Message ID (reassembly of multiple msgs) 44 Byte user payload (~84% efficient) LEN: Length of data in this segment CRC: bit segment CRC Each class of service has A different header format (in addition to the 5 byte ATM header) USER PDU (DLC or NL) ATM CELL ATM CELL AAL 3/4 allows multiplexing, reliability, & error detection but is fairly complex to process and adds much overhead AAL 5 was introduced to support IP traffic Fewer functions but much less overhead and complexity Slide 57 Slide 58 ATM cell switches ATM summary ATM is mostly used as a core network technology Input Cell Processing Output ATM Advantages Cell Input 2 Processing Output 2 Switch Fabric Ability to provide QoS Ability to do traffic management Fast cell switching using relatively short VC numbers Input m Cell Processing Input Q's Design issues Input vs. output queueing Head of line blocking Fabric speed Control S/W Output Q's Output m ATM disadvantages It not IP - most everything was design for TCP/IP It s not naturally an end-to-end protocol Does not work well in heterogeneous environment Was not design to inter-operate with other protocols Not a good match for certain physical media (e.g., wireless) Many of the benefits of ATM can be borrowed by IP Cell switching core routers Label switching mechanisms Slide 59 Slide 6

16 Multi-Protocol Label Switching (MPLS) Label Switching As more services with fixed throughput and delay requirements become more common, IP will need virtual circuits (although it will probably call them something else) RG, April 28, 994 Router makers realize that in order to increase the speed and capacity they need to adopt a mechanism similar to ATM Switch based on a simple tag not requiring complex routing table look-ups Use virtual circuits to manage the traffic (QoS) Use cell switching at the core of the router First attempt: IP-switching Routers attempt to identify flows Define a flow based on observing a number of packets between a given source and destination (e.g., 5 packets within a second) Map IP source-destination pairs to ATM VC s Distributed algorithm where each router makes its own decision Multi-protocol label switching (MPLS) Also known as Tag switching Does not depend on ATM Add a tag to each packet to serve as a VC number Tags can be assigned permanently to certain paths Slide 6 Slide 62 Label switching can be used to create a virtual mesh with the core network References Routers at the edge of the core network can be connected to each other using labels Packets arriving at an edge router can be tagged with the label to the destination edge router Core network D TCP/IP Illustrated (Vols. &2), Stevens Computer Networks, 4th ed., Tenenbaum Tunneling Significantly simplifies routing in the core Computer networking, 2nd ed., Kurose and Ross Interior routers need not remember all IP prefixes of outside world D Allows for traffic engineering Assign capacity to labels based on demand Label switched routes Slide 63 Slide 64

17 Slide 65 7 bits 24 bits Class A netid hostid 4 bits 6 bits Class B netid hostid 2 bits 8 bits Class C netid hostid 28 bits Class D netid hostid 27 bits Class E (reserved for future use)